Loading...

Predicting Wind Turbine Failures using Machine Learning

Ahmadi, Mohsen | 2024

0 Viewed
  1. Type of Document: M.Sc. Thesis
  2. Language: Farsi
  3. Document No: 57464 (05)
  4. University: Sharif University of Technology
  5. Department: Electrical Engineering
  6. Advisor(s): Abasspour Tehranifard, Ali; Fatemizadeh, Emad
  7. Abstract:
  8. Nowadays, the development of renewable energy sources is in line with the increasing capacity and installation of wind energy resources. Wind turbines, due to their installation in harsh areas, have high maintenance and repair costs, which is one of the main challenges for their further development. Given the high costs of repairs, predicting faults and planning for optimal maintenance can optimize performance and extend the useful life of the turbine. The aim of this research is to provide an approach to predict faults in wind turbines using data from the control and monitoring system. This data includes various parameters such as wind speed, output power, temperature of the generator bearings and gearbox, and generator windings, which are used to analyze and predict the health status of the turbine. The proposed method in this research consists of several steps: In the first step, the normal behavior of the wind turbine must be modeled. This is done using a two-step approach involving the removal of outliers in each of the input parameters and the removal of outliers resulting from non-nominal power generation. A density-based clustering algorithm is used to remove these values. After this step, the cleaned data is scaled to a uniform range and the data structure is modified to eliminate the time sequence. The data is then applied to various machine learning models, and their performance is compared. At this stage, the final model learns the normal behavior of the turbine using the training data. Evaluation data is applied to the model, and the deviation between the actual and predicted values is assessed. To detect temperature deviation trends, a weighted exponential moving average is used. If the values exceed a specified threshold, they are considered as an early warning of failure. To prevent false alarms, in addition to the magnitude of the predicted deviation, the persistence of this condition is also considered as a supplementary indicator. The proposed approach is validated using a real database from a wind farm in South Africa consisting of 5 wind turbines
  9. Keywords:
  10. Wind Turbine ; Machine Learning ; Preventive Maintenance ; Defect Prediction ; Performance Optimization ; Wind Turbine Cost

 Digital Object List

 Bookmark

...see more